Leveraging Generative Pre-trained Transformers for the Integration of Environmental, Social, and Governance Considerations into Investment Management for Thai Stock

The problem

Thai 56-1 One Reports are long and complex, while sustainability information is often treated as non-financial and is difficult for general investors to combine with financial analysis. Investors therefore need an accessible way to extract material ESG issues without reading hundreds of pages unaided.

Who benefits:

• Thai equity analysts and portfolio managers
• ESG research and stewardship teams
• Investment-compliance and risk professionals
• Regulators and financial-technology developers

Method

The software grounds a GPT in SET industry-specific ESG Metrics guidance and 2021 ESG reports from all sectors. A Python and LangChain workflow combines a prompt chain, which assigns the role of an experienced ESG analyst, with a retrieval toolchain backed by document embeddings and a vector database. Users upload a Thai 56-1 One Report and receive three structured outputs: Materiality Assessment, Stakeholder Engagement, and Challenges and Limitations. The paper illustrates the workflow with the 454-page PTT and 350-page SCBX reports.

Results

The examples show that the GPT can distil lengthy One Reports into concise ESG points and support materiality review and engagement planning. The study does not benchmark accuracy against other methods, and the system is not trained for real-time stock prediction or definitive investment advice. Data bias, non-standard reporting, incomplete public information, and changing ESG issues remain important risks; the authors recommend broader data, advanced validation, continuous updates, and human oversight.